
from tensorflow.keras.utils import set_random_seed, plot_model
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense, Dropout, Conv2D, Flatten, MaxPool2D
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
from tensorflow.keras.applications import VGG16, VGG19
from sklearn.model_selection import train_test_split
from sklearn.cluster import KMeans
from sklearn.preprocessing import OneHotEncoder, LabelEncoder
from PIL import Image
import matplotlib.pyplot as plt
import PIL.Image as pilimg
import numpy as np
import os
import cv2


flowers=['daisy', 'dandelion', 'rose','sunflower', 'tulip']


def get_image(flower):
    # 이미지 담을 리스트, 파일 경로 생성
    hr_image = []
    target=[]
    
    for flower in flowers:
        a=1
        PATH = r'./file/'+flower+'/'

        # 폴더 내 모든 이미지 읽고 저장
        for filename in os.listdir(PATH): 
            if a<=500:
                img = cv2.imread(PATH + filename)
                # 색깔이 바뀌어서 나오는 것 바로잡기
                img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
                resize_img=cv2.resize(img, (224, 224))
                hr_image.append(resize_img)

                target.append(flower)
                a+=1
                

    # numpy 배열 형태로 저장
    hr_image = np.array(hr_image)
    target=np.array(target)
    
    return hr_image, target


data, target=get_image(flowers)


data.shape

(2500, 224, 224, 3)


target.shape

(2500,)


# 스케일링
data=data/255.0


plt.imshow(data[500])

<matplotlib.image.AxesImage at 0x200a19b8670>


x_train, x_test, y_train, y_test=train_test_split(
data, target, stratify=target, test_size=0.2, random_state=42)


print(f'x_train.shape : {x_train.shape}, x_test.shape : {x_test.shape}')
print(f'y_train.shape : {y_train.shape}, y_test.shape : {y_test.shape}')

x_train.shape : (2000, 224, 224, 3), x_test.shape : (500, 224, 224, 3)
y_train.shape : (2000,), y_test.shape : (500,)


x_train, x_val, y_train, y_val=train_test_split(
x_train, y_train, stratify=y_train, test_size=0.2, random_state=42)


print(f'x_train.shape : {x_train.shape}, x_val.shape : {x_val.shape}')
print(f'y_train.shape : {y_train.shape}, y_val.shape : {y_val.shape}')

x_train.shape : (1600, 224, 224, 3), x_val.shape : (400, 224, 224, 3)
y_train.shape : (1600,), y_val.shape : (400,)


y_train=y_train.reshape(-1, 1)
y_test=y_test.reshape(-1, 1)
y_val=y_val.reshape(-1, 1)


ohe=OneHotEncoder()


ohe.fit(y_train)

OneHotEncoder()


y_train=ohe.transform(y_train)
y_test=ohe.transform(y_test)
y_val=ohe.transform(y_val)


y_train=y_train.toarray()
y_test=y_test.toarray()
y_val=y_val.toarray()


y_train

array([[0., 1., 0., 0., 0.],
       [0., 0., 1., 0., 0.],
       [1., 0., 0., 0., 0.],
       ...,
       [0., 0., 0., 0., 1.],
       [0., 1., 0., 0., 0.],
       [0., 0., 0., 0., 1.]])


# W, b 초기값 고정
set_random_seed(11)


def cnnModel(num):
    model=Sequential()
    model.add(Conv2D(filters=num, kernel_size=3,
                    padding='same',
                    input_shape=(224, 224, 3)))
    model.add(MaxPool2D())
    model.add(Conv2D(filters=num*2, kernel_size=3,
                    padding='same'))
    model.add(Conv2D(filters=num*2, kernel_size=3,
                    padding='same'))
    model.add(MaxPool2D())
    model.add(Dropout(0.3))
    model.add(Flatten())
    model.add(Dense(100, activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(100, activation='relu'))
    model.add(Dense(5, activation='softmax'))
    model.compile(loss='categorical_crossentropy', optimizer='adam',
                 metrics=['accuracy'])
    
    return model


cnn_model=cnnModel(64)


stopCB=EarlyStopping(monitor='val_accuracy', patience=5)


MODEL_PATH='./model/model.{epoch:02d}-{val_loss:.2f}.h5'
modelCB=ModelCheckpoint(filepath=MODEL_PATH,
                       save_best_only=True)


his=cnn_model.fit(x_train, y_train, epochs=10, batch_size=100,
             validation_data=(x_val, y_val),
             callbacks=[stopCB, modelCB])

Epoch 1/10
16/16 [==============================] - 291s 18s/step - loss: 8.0711 - accuracy: 0.2381 - val_loss: 1.6439 - val_accuracy: 0.2850
Epoch 2/10
16/16 [==============================] - 259s 16s/step - loss: 1.6289 - accuracy: 0.2587 - val_loss: 1.5239 - val_accuracy: 0.2725
Epoch 3/10
16/16 [==============================] - 260s 16s/step - loss: 1.5701 - accuracy: 0.2237 - val_loss: 1.5203 - val_accuracy: 0.2850
Epoch 4/10
16/16 [==============================] - 259s 16s/step - loss: 1.5749 - accuracy: 0.2250 - val_loss: 1.5543 - val_accuracy: 0.2475
Epoch 5/10
16/16 [==============================] - 261s 16s/step - loss: 1.5600 - accuracy: 0.2375 - val_loss: 1.5279 - val_accuracy: 0.2725
Epoch 6/10
16/16 [==============================] - 262s 17s/step - loss: 1.5554 - accuracy: 0.2594 - val_loss: 1.5086 - val_accuracy: 0.3300
Epoch 7/10
16/16 [==============================] - 261s 16s/step - loss: 1.5290 - accuracy: 0.2619 - val_loss: 1.5202 - val_accuracy: 0.2700
Epoch 8/10
16/16 [==============================] - 262s 16s/step - loss: 1.4954 - accuracy: 0.3031 - val_loss: 1.4558 - val_accuracy: 0.3375
Epoch 9/10
16/16 [==============================] - 262s 16s/step - loss: 1.4800 - accuracy: 0.3081 - val_loss: 1.4462 - val_accuracy: 0.3425
Epoch 10/10
16/16 [==============================] - 261s 16s/step - loss: 1.4798 - accuracy: 0.3150 - val_loss: 1.4554 - val_accuracy: 0.4150


model=VGG16(input_shape=(224, 224, 3), include_top=False, weights='imagenet')

for layer in vgg_model.layers[:19]:
        layer.trainable = False


model = Sequential([
        model,
        Conv2D(64, kernel_size=3, padding='same'),
        MaxPool2D(),
        Dropout(0.5),
        Flatten(),
        Dense(5 , activation='softmax')])
model.compile(optimizer = "adam" , loss = 'categorical_crossentropy' , metrics = ['accuracy'])
model.summary()

Model: "sequential_12"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 vgg16 (Functional)          (None, 7, 7, 512)         14714688  
                                                                 
 conv2d_28 (Conv2D)          (None, 7, 7, 64)          294976    
                                                                 
 max_pooling2d_24 (MaxPoolin  (None, 3, 3, 64)         0         
 g2D)                                                            
                                                                 
 dropout_23 (Dropout)        (None, 3, 3, 64)          0         
                                                                 
 flatten_12 (Flatten)        (None, 576)               0         
                                                                 
 dense_19 (Dense)            (None, 5)                 2885      
                                                                 
=================================================================
Total params: 15,012,549
Trainable params: 297,861
Non-trainable params: 14,714,688
_________________________________________________________________


history = model.fit(x_train,y_train, batch_size = 100 , epochs = 12 , validation_data = (x_val, y_val), callbacks = [stopCB, modelCB])

Epoch 1/12
16/16 [==============================] - 399s 25s/step - loss: 1.4943 - accuracy: 0.4538 - val_loss: 0.8519 - val_accuracy: 0.7175
Epoch 2/12
16/16 [==============================] - 405s 26s/step - loss: 0.7115 - accuracy: 0.7431 - val_loss: 0.6840 - val_accuracy: 0.7850
Epoch 3/12
16/16 [==============================] - 399s 25s/step - loss: 0.4990 - accuracy: 0.8350 - val_loss: 0.6386 - val_accuracy: 0.7900
Epoch 4/12
16/16 [==============================] - 385s 24s/step - loss: 0.3715 - accuracy: 0.8756 - val_loss: 0.6391 - val_accuracy: 0.7850
Epoch 5/12
16/16 [==============================] - 381s 24s/step - loss: 0.3321 - accuracy: 0.8888 - val_loss: 0.6547 - val_accuracy: 0.7650
Epoch 6/12
16/16 [==============================] - 378s 24s/step - loss: 0.2576 - accuracy: 0.9194 - val_loss: 0.5872 - val_accuracy: 0.8075
Epoch 7/12
16/16 [==============================] - 377s 24s/step - loss: 0.2064 - accuracy: 0.9438 - val_loss: 0.5709 - val_accuracy: 0.8025
Epoch 8/12
16/16 [==============================] - 378s 24s/step - loss: 0.1649 - accuracy: 0.9594 - val_loss: 0.5760 - val_accuracy: 0.8050
Epoch 9/12
16/16 [==============================] - 379s 24s/step - loss: 0.1510 - accuracy: 0.9544 - val_loss: 0.5880 - val_accuracy: 0.8275
Epoch 10/12
16/16 [==============================] - 378s 24s/step - loss: 0.1202 - accuracy: 0.9719 - val_loss: 0.5714 - val_accuracy: 0.8200
Epoch 11/12
16/16 [==============================] - 375s 24s/step - loss: 0.1020 - accuracy: 0.9806 - val_loss: 0.5674 - val_accuracy: 0.8200
Epoch 12/12
16/16 [==============================] - 375s 24s/step - loss: 0.0799 - accuracy: 0.9862 - val_loss: 0.5702 - val_accuracy: 0.8350


plt.plot(history.epoch, history.history['accuracy'])
plt.plot(history.epoch, history.history['val_accuracy'])
plt.legend(['accuracy', 'val_accuracy'])
plt.grid()
plt.show()


model.evaluate(x_test, y_test)

16/16 [==============================] - 83s 5s/step - loss: 0.5145 - accuracy: 0.8380

[0.5144973993301392, 0.8379999995231628]


np.round(model.predict(x_test[0].reshape(-1, 224, 224, 3)))

1/1 [==============================] - 0s 231ms/step

array([[0., 0., 0., 0., 1.]], dtype=float32)


y_test[0]

array([0., 0., 0., 0., 1.])


# 폴더 내 모든 이미지 읽고 저장
pre_flower=[]
PATH = './new_img/'

for filename in os.listdir(PATH):
    image=cv2.imread(PATH+filename)
    image=cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    image=cv2.resize(image, (224, 224))
    pre_flower.append(image)
    
pre_flower=np.array(pre_flower)
pre_flower=pre_flower/255.0


def pre_label(pre_target):
    '''
    예측값의 라벨 출력
    '''
    
    if pre_target[0][0]==1: return 'daisy'
    elif pre_target[0][1]==1: return 'dandelion'
    elif pre_target[0][2]==1: return 'rose'
    elif pre_target[0][3]==1: return 'sunflower'
    else: return 'tulip'


def pre_img(i, pre_target, answers):
    plt.imshow(pre_flower[i])
    plt.show()
    
    print(f'예측 => {pre_target} {pre_label(pre_target)}')
    print(f'정답 => {answers[i]}')


def model_predict(model):
    for i in range(len(pre_flower)):
        answers=['daisy','daisy', 'daisy', 'daisy', 'daisy',
                 'dandelion', 'dandelion', 'dandelion', 
                 'rose', 'rose', 
                 'sunflower', 'sunflower', 
                 'tulip', 'tulip', 'tulip', 'tulip']
        pre_target=np.round(model.predict(pre_flower[i].reshape(-1, 224, 224, 3)))

        if i in [0, 1, 2, 3, 4]:
            pre_img(i, pre_target, answers)
            if int(pre_target[0][0])==1: print('✿ڿڰۣ 데이지 정답🌼 ✿ڿڰۣ')
            else: print('땡 ( ꃼ-ꃼ )')

        elif i in [5, 6, 7]:
            pre_img(i, pre_target, answers)

            if int(pre_target[0][1])==1: print('✿ڿڰۣ 민들레 정답🌱 ✿ڿڰۣ')
            else: print('땡 ( ꃼ-ꃼ )')

        elif i in [8, 9]:
            pre_img(i, pre_target, answers)

            if int(pre_target[0][2])==1: print('✿ڿڰۣ 장미 정답🌹 ✿ڿڰۣ')
            else: print('땡 ( ꃼ-ꃼ )')

        elif i in [10, 11]:
            pre_img(i, pre_target, answers)

            if int(pre_target[0][3])==1: print('✿ڿڰۣ 해바라기 정답🌻 ✿ڿڰۣ')
            else: print('땡 ( ꃼ-ꃼ )')

        else:
            pre_img(i, pre_target, answers)

            if int(pre_target[0][4])==1: print('✿ڿڰۣ 튤립 정답🌷 ✿ڿڰۣ')
            else: print('땡 ( ꃼ-ꃼ )')


model_predict(cnn_model)

1/1 [==============================] - 0s 78ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => daisy
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 76ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => daisy
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 71ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => daisy
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 72ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => daisy
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 72ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => daisy
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 74ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => dandelion
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 70ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => dandelion
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 75ms/step

예측 => [[0. 0. 0. 0. 0.]] tulip
정답 => dandelion
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 78ms/step

예측 => [[0. 0. 1. 0. 0.]] rose
정답 => rose
✿ڿڰۣ 장미 정답🌹 ✿ڿڰۣ
1/1 [==============================] - 0s 83ms/step


model_predict(model)

1/1 [==============================] - 0s 256ms/step

예측 => [[1. 0. 0. 0. 0.]] daisy
정답 => daisy
✿ڿڰۣ 데이지 정답🌼 ✿ڿڰۣ
1/1 [==============================] - 0s 249ms/step

예측 => [[1. 0. 0. 0. 0.]] daisy
정답 => daisy
✿ڿڰۣ 데이지 정답🌼 ✿ڿڰۣ
1/1 [==============================] - 0s 240ms/step

예측 => [[1. 0. 0. 0. 0.]] daisy
정답 => daisy
✿ڿڰۣ 데이지 정답🌼 ✿ڿڰۣ
1/1 [==============================] - 0s 208ms/step

예측 => [[1. 0. 0. 0. 0.]] daisy
정답 => daisy
✿ڿڰۣ 데이지 정답🌼 ✿ڿڰۣ
1/1 [==============================] - 0s 201ms/step

예측 => [[1. 0. 0. 0. 0.]] daisy
정답 => daisy
✿ڿڰۣ 데이지 정답🌼 ✿ڿڰۣ
1/1 [==============================] - 0s 191ms/step

예측 => [[0. 1. 0. 0. 0.]] dandelion
정답 => dandelion
✿ڿڰۣ 민들레 정답🌱 ✿ڿڰۣ
1/1 [==============================] - 0s 189ms/step

예측 => [[0. 1. 0. 0. 0.]] dandelion
정답 => dandelion
✿ڿڰۣ 민들레 정답🌱 ✿ڿڰۣ
1/1 [==============================] - 0s 199ms/step

예측 => [[0. 0. 0. 0. 1.]] tulip
정답 => dandelion
땡 ( ꃼ-ꃼ )
1/1 [==============================] - 0s 180ms/step

예측 => [[0. 0. 1. 0. 0.]] rose
정답 => rose
✿ڿڰۣ 장미 정답🌹 ✿ڿڰۣ
1/1 [==============================] - 0s 192ms/step

다섯가지 꽃 종류 분류 🌼🌱🌹🌻🌷¶

[1] 데이터 불러오기¶

[2] 데이터 분리¶

target OneHotEncoding¶

[3] 모델 생성 및 학습¶

CNN¶

VGG16¶

[4] 예측¶

CNN 모델 예측¶

VGG16 모델 예측¶